Why AI Experts Say the Real Danger Is a Stampede, Not a Race

Imagine a packed building with one exit. If everyone walks out calmly, no one gets hurt. But if a few people push, panic spreads, and the crowd can crush itself at the doorway — even though the door was never the problem. That's the analogy AI researcher Richard Ngo wants us to use for artificial intelligence, instead of the familiar phrase "AI arms race." He argues the race framing makes AI companies paranoid and hostile toward each other, which is exactly the mindset that causes a stampede.

Ngo borrows the idea from sociologist James Coleman, who described evacuations as a real-life version of the prisoner's dilemma — a classic situation where everyone would do better by cooperating, but each person is tempted to defect for a personal edge. In a stampede, the single smartest move is to calm the crowd, not to shove harder. Ngo also dislikes that "AI race" implies there will be a winner, when the risk is that everybody loses.

So where are we now? Ngo says the crowd is still jostling as it moves, but almost nobody has been seriously hurt. There is still time to defuse the herd mentality. He notes that many real stampedes were not triggered by a fire but by panic itself — the Hillsborough disaster, the Mecca tunnel tragedy, a temple stampede in India — which he finds closer to the AI situation, because nobody agrees on exactly what danger we're fleeing.

He traces the history: early transhumanists like Ray Kurzweil shouted about a fire nobody else noticed, while thinkers like Nick Bostrom and Eliezer Yudkowsky warned that the exit door might be jammed shut. Today's AI leaders, he notes, walked toward that door talking about how wonderful the outside would be. The takeaway for the rest of us: the tone AI companies set — competitive or cooperative — is not just corporate culture. It shapes whether this ends in an orderly exit or a crush.